KEY POINTS
Tech giants are set to spend $700 billion on AI build-outs in 2026 alone, and every dollar of that spend needs a power source behind it.
Wells Fargo projects AI power demand surges 550% by 2026 — from 8 TWh to 52 TWh — and another 1,150% to 652 TWh by 2030. That's not a trend. That's a structural shift.
Four energy categories — nuclear, natural gas, regulated utilities, and fuel cells — each offer a different risk-reward profile for investors willing to look beyond the usual AI trade.
TOP STORY
AI Needs More Power Than the U.S. Grid Can Deliver
Here's something most people miss about the AI boom. While everyone's been watching Nvidia and the chip stocks, a quieter race has been heating up in a less glamorous corner of the market.
Power. Plain old electricity.
A single ChatGPT query uses about 10 times more electricity than a regular Google search. Now multiply that by billions of queries a day, plus all the model training, cloud storage, and real-time inference happening across hundreds of new data centers. The numbers get big, fast.
Meta, Microsoft, Amazon, and Alphabet alone are projected to spend $700 billion on AI infrastructure in 2026. The global data center sector is growing at a 14% CAGR through 2030, requiring up to $3 trillion in total investment. And Morgan Stanley Research has flagged a potential shortfall of 49 GW in available power access by 2028 against 74 GW in demand.
That gap matters more than you might think. Because whoever fills it gets paid — consistently, under long-term contracts, with pricing power. That's a pretty good deal for investors willing to look past the obvious AI plays.
WHY IT MATTERS TO YOU
Why "Speed to Power" Is Now the Most Important Phrase in Infrastructure
Think of it this way. You're building a giant data center. You have the land. You have the permits. You have the servers. But if you can't get reliable, affordable electricity hooked up quickly, nothing works. You're running at a loss from day one.
That's exactly the bottleneck companies like Microsoft, Amazon, and Google are hitting right now. Grid interconnection queues in some states stretch 5 to 7 years. The U.S. power grid — most of it built between the 1950s and 1970s — wasn't designed for this kind of load. As Clift Pompee, VP of Power at Compass Datacenters, put it recently: "2026 is a pivotal year for the future of the U.S. power grid. Approximately 70% of the grid is approaching end of its life cycle."
Expert Voice
Morgan Stanley Research forecasts U.S. data center demand could reach 74 GW by 2028, with a projected shortfall of about 49 GW in available power access. Large tech companies are likely to commit more than $1 trillion of spending in just the 2025–2026 period. Strategic financing will be a critical enabler. — Morgan Stanley, February 2026
So what happens when demand is exploding and supply is constrained? Energy companies with existing capacity, long-term contracts, and proximity to data center hubs get to write their own ticket. That's what we're watching play out right now.

BY THE NUMBERS
The Scale Is Hard to Wrap Your Head Around
$700B — projected hyperscaler AI capex in 2026 alone (Meta, Microsoft, Amazon, Alphabet)
550% — projected surge in AI power demand by 2026 vs. 2024 (Wells Fargo)
14% CAGR — global data center sector growth through 2030, requiring up to $3 trillion total
47.1 GW — power requests Dominion Energy has on file from data centers in Virginia alone
$20B — Bloom Energy's total current backlog, including $6B in product and $14B in service
$7.3B — BWX Technologies' backlog at end of 2025, up 50% year-over-year
15 GW — powered data center hubs NextEra Energy is targeting by 2035
6 consecutive years — U.S. natural gas production is set to hit record levels in 2026 for the sixth year in a row
THE PLAYS
Four Ways to Get Positioned — Ranked by Time Horizon
☢️ Nuclear — The Long Game (3–10 Years)
Constellation Energy (CEG) is the cleanest pure-play nuclear name. It's already locked up 20-year power purchase agreements with both Meta Platforms and Microsoft. Nuclear gives data centers what renewables can't: steady, 24/7 baseload power with no major emissions. GE Vernova (GEV) adds an infrastructure angle through its BWRX-300 small modular reactor, with the first deployment scheduled in Canada in 2029. And BWX Technologies (BWXT) plays the components side — its backlog hit $7.3 billion at end of 2025, up 50% year-over-year. These aren't short-term trades. They're multi-decade infrastructure bets.
🔥 Natural Gas — The Near-Term Bridge (Now–5 Years)
Energy Transfer (ET) is the most interesting near-term name. It's already signed direct data center supply agreements with Oracle, CloudBurst, and Fermi America — and trades at a forward P/E of roughly 11.4x, which is cheap relative to the opportunity. Enbridge (ENB) offers a hybrid angle (pipeline + renewable) with a 5.2% dividend yield that pays you to wait. Archrock (AROC) is more niche — it provides natural gas compression services and is planning $250–$275 million in capex to grow with the data center boom and LNG export demand. Natural gas turbines are the practical answer right now because the grid simply can't add enough renewable capacity fast enough.
🏭 Regulated Utilities — Steady Capital at Scale (2–7 Years)
Virginia is the world's #1 data center market, and Dominion Energy (D) is its primary power provider — with $50 billion in capital investment planned from 2025 through 2029 and 47.1 GW of power requests already in the pipeline. Entergy (ETR) is building two new gas-fired plants totaling 1.5 GW specifically to power Meta's $10 billion data center in Northeast Louisiana, and plans to invest $41 billion between 2026 and 2029. NextEra Energy (NEE) is the largest utility in the world by market cap and is partnering with Google on GW-scale data center campuses and Exxon on a 1.2 GW gas plant — while targeting 15 GW of powered data center hubs by 2035. These are slower-moving names, but the earnings visibility is exceptional.
🔋 Fuel Cells — Fastest to Power, Highest Risk (Now — But Watch the Valuation)
Bloom Energy (BE) is the most talked-about name in the space right now — and for good reason. It posted record revenue of $2.02 billion in 2025 (+37.3% YoY), guided for 58% growth to $3.1–$3.3 billion in 2026, and has a $20 billion backlog. Oracle expanded its procurement deal to up to 2.8 GW of Bloom's fuel cell systems. The key advantage: Bloom can deploy on-site in around 90 days — faster than any grid-connected alternative. But the stock carries a forward P/E of roughly 104x. It's a high-conviction, high-risk position. Entry point matters a lot here.
FACT-CHECK VERIFIED
Key Energy Plays — Comparative Data

WHAT TO WATCH
Three Things That Could Move This Trade
The GRID Act
New legislation aims to prevent data-center cost increases from hitting residential utility bills. If it passes with teeth, it creates real earnings uncertainty for utilities like Dominion and Entergy with heavy data center exposure in their service territories.
SMR Timeline Risk
Small modular reactors like GE Vernova's BWRX-300 aren't scheduled for deployment until 2029 at the earliest. Any permitting delays, cost overruns, or supply chain hiccups could push that out — and the market will reprice accordingly.
Natural Gas Pricing
Record U.S. production should keep gas prices manageable — which is great for data center operators but caps upside for pure-play gas stocks. The volume story still works; the price story is less clear.
THE BOTTOM LINE
The AI trade isn't just semiconductors and software.
Power is the constraint — and the companies solving it are already signing 20-year contracts with the biggest tech firms on the planet.
If you believe AI infrastructure spending is real and durable, then the energy buildout behind it deserves a serious look.
Just know your timeline.
Bloom Energy is a now-bet.
Nuclear is a decade-bet. Utilities are somewhere in between.
Entry point and patience both matter here — this is a multi-year trade, not a momentum play.



